BlueBench - RAG General: leaderboard
BlueBench retrieval-augmented generation task using CLAP NQ-style long-form question answering with grounded passages.
Metric: Score (%). Source: huggingface.co. 18 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4.1 Mini | 54.58 | #355 |
| 2 | Mistral Medium 3 | 52.57 | #500 |
| 3 | Granite 3.3 8B Instruct | 51.92 | #825 |
| 4 | Mistral Large | 51.08 | #710 |
| 5 | GPT-4.1 Nano | 50.88 | #731 |
| 6 | Llama 3.3 70B Instruct | 49.02 | #534 |
| 7 | GPT-4o | 48.9 | #343 |
| 8 | GPT-4.1 | 48.46 | #248 |
| 9 | Pixtral-12B | 47.85 | #808 |
| 10 | Llama 3.2 3B Instruct | 47.81 | #1352 |
| 11 | Llama 3.2 1B Instruct | 44.7 | #1539 |
| 12 | O3 Mini | 44.06 | #272 |
| 13 | O1 | 41.24 | #183 |
| 14 | O4 Mini | 40.32 | #180 |
Interactive version: theaggregate.ai/benchmark?slug=bluebench-rag-general · How It Works · Data refreshed daily, snapshot 2026-10-20.